Trc

DC

a

ARRAA

KFEEESSR

1

lmlbahcnotac

(

h0

Energy and Buildings 143 (2017) 25–34

Contents lists available at ScienceDirect

Energy and Buildings

journa l homepage: www.e lsev ier .com/ locate /enbui ld

he intersection of energy and justice: Modeling the spatial,acial/ethnic and socioeconomic patterns of urban residential heatingonsumption and efficiency in Detroit, Michigan

ominic J. Bednar ∗, Tony Gerard Reames, Gregory A. Keoleianenter for Sustainable Systems, School of Natural Resources and Environment, University of Michigan, 440 Church St., Ann Arbor, MI 48109, United States

r t i c l e i n f o

rticle history:eceived 18 August 2016eceived in revised form 19 February 2017ccepted 9 March 2017vailable online 12 March 2017

eywords:uel povertynergy justicenergy consumptionnergy efficiencypatial analysis

a b s t r a c t

Residential energy conservation and efficiency programs are strategic interventions to reduce consump-tion and increase affordability. However, the inability to identify and distinguish between high energyconsumers and highly energy inefficient households has led to ineffective program targeting. Addition-ally, little is known about the spatial, racial and socioeconomic patterns of urban residential energyconsumption and efficiency. Publicly available data from the U.S. Energy Information Administrationand the U.S. Census Bureau are used with bottom-up modeling and small-area estimation techniques topredict mean annual heating consumption and energy use intensity (EUI), an energy efficiency proxy, atthe census block group level in Detroit (Wayne County), Michigan. Using geographic information sys-tems, results illustrate spatial disparities in energy consumption and EUI. Bivariate analysis show nostatistical relationship between race/ethnicity and energy consumption; however, EUI is correlated withracial/ethnic makeup; percent White (−0.28), African American (0.24) and Hispanic (0.16). Income and

pace heatingesidential buildings

housing tenure reveal inverse relationships with consumption and efficiency. Though areas with highermedian incomes and homeownership exhibited higher consumption (0.28 and 0.56, respectively), theyhad lower EUIs (−0.48 and −0.38, respectively). This study provides evidence supporting approaches forconservation and energy efficiency program targeting that recognizes the significance of race, ethnicity,place and class.

© 2017 Elsevier B.V. All rights reserved.

find distinguishable spatial disparities in both consumption and

. Introduction

Residential utility costs place a disproportionate burden onow-income households. Following the Great Recession, nearly 14

illion American households had utility bills in arrears and 2.2 mil-ion households experienced utility shutoffs [1]. Residential energyurdens, or the percentage of annual income spent on energy costsre a major source of utility hardship. While the average Americanousehold spends 7.2% of their annual income on residential energyosts, the average low-income household has an energy burdenearly double, spending 13.8% [2]. Energy burden disparities areften expressed through the concept of fuel poverty, also referred

o as energy insecurity [3,4]. Fuel poverty reflects an inability of

household to meet basic energy needs or to adequately heat orool their home [3]. Fuel poverty results from the interplay between

∗ Corresponding author.E-mail addresses: dbednar@umich.edu (D.J. Bednar), treames@umich.edu

T.G. Reames), gregak@umich.edu (G.A. Keoleian).

ttp://dx.doi.org/10.1016/j.enbuild.2017.03.028378-7788/© 2017 Elsevier B.V. All rights reserved.

low household incomes, rising energy costs and energy inefficienthomes [3].

Amid solutions to alleviate fuel poverty, energy conservationand efficiency retrofit programs have proven successful [5–8].However, the inability to identify and distinguish between house-holds with high energy consumption compared to those thatare highly energy inefficient has halted interventions from utiliz-ing systematic approaches to appropriately and effectively targetenergy conservation and efficiency programs.

The need for more effective targeting is supported by previousstudies exploring the spatial dynamics of energy consumption that

energy use intensity (EUI).1 For instance, Heiple and Sailor [9] usingnational data, building energy simulation and geospatial modeling

1 According to the U.S. Department of Energy, “Declines in energy intensity area proxy for efficiency improvements, provided a) energy intensity is representedat an appropriate level of disaggregation to provide meaningful interpretation, andb) other explanatory and behavioral factors are isolated and accounted for” (DOEa)[52].

2 nd Bu

taetactrtttpbauetakb

ianrbcamhtrmatgco

2

settiretota[vhc

asepcsr

Detroit has long been the most segregated metropolitan areain the nation, having a majority African American and Hispaniccity population and a majority White suburban population [35].This segregation is evident in Fig. 3, a dot density map illustrating

2 The Pew Research Center developed a single Residential Income SegregationIndex (RISI) score for the nation’s top 30 metropolitan areas. The score is cal-

6 D.J. Bednar et al. / Energy a

echniques found variations in peak energy profiles for electricitynd natural gas across building types in Houston, Texas. Howardt al. [10] built models from citywide data to estimate building sec-or EUI finding major differences in the magnitude of consumptionnd spatial variation across New York City. Santamouris et al. [11]onducted interviews on household and housing unit characteris-ics finding higher costs per person and unit area for low-incomeesidence in Athens. These studies provide rich information onhe relationship between place and energy consumption; however,heir focus on both commercial and residential energy consump-ion makes it difficult to identify residential energy disparities forrogram targeting. Moreover, few studies investigate correlationsetween residential energy consumption, efficiency, race/ethnicitynd socioeconomic status for a more holistic understanding ofrban residential energy dynamics. Reames [12] developed a modelstimating urban residential heating EUI and found positive rela-ionships with areas with higher percentages of racial minoritiesnd lower socioeconomics. Albeit some exploration, little remainsnown about the spatial, racial and socioeconomic differencesetween residential energy consumption and efficiency.

To this end, this paper develops models for residential heat-ng consumption and efficiency at the census block group levelnd explores the spatial patterns alongside racial and socioeco-omic relationships in Detroit (Wayne County), Michigan. Theemainder of this paper is structured as follows. Section 2 presentsackground information on modeling energy consumption, effi-iency and disparities. Section 3 describes the study area, datand methodological framework for first developing two regressionodels to estimate residential heating energy consumption and

eating EUI, then secondly, using small area estimation techniqueso predict consumption and EUI in the study area. Section 4 presentsesults of the regression models, spatial distributions of results

apped using geographic information systems (GIS) and bivariatenalysis of the relationship between predicted energy consump-ion and efficiency with selected racial and socioeconomic blockroup characteristics. Section 5 discusses key results, policy impli-ations and study limitations. Lastly, concluding remarks and areasf future research are presented in Section 6.

. Background

To understand the factors that impact energy consumption,cholars apply two general frameworks: the physical-technical-conomic model (PTEM) and the lifestyle and social-behavioralradition (LSB) [13–23]. In 1993, Lutzenhiser proposed the PTEMradition arguing that the physical characteristics of buildings,nvestment in technical energy efficiency, energy prices and envi-onmental factors are integral to understanding and managingnergy consumption. On the other hand, the LSB tradition con-ends that these factors alone can only offer minimal explanationf energy consumption in the built environment and draws atten-ion to the importance of human occupants of the building, suchs, social (noneconomic), behavioral, cultural and lifestyle factors13,14,17–20,24,25]. The models developed for this study includeariables merging the PTEM and LSB modeling traditions for a moreolistic understanding of residential energy consumption and effi-iency.

Individual housing unit energy data is often not readily avail-ble for exploring residential energy dynamics at various spatialcales. Thus, the absence of detailed information on residentialnergy use presents an impediment to spatially identifying fuel

oor households and developing strategic conservation and effi-iency program targeting. As a result, scholars have employedmall area estimation statistical techniques to spatially exploreesidential energy patterns. This approach requires finding the

ildings 143 (2017) 25–34

best predictors to model energy consumption and efficiency, forinstance, energy characteristics of housing structure and a selectionof householder characteristics; then, connect to matching spatialdata (i.e. census data).

A growing body of literature investigating geographicalapproaches to target fuel poverty in Europe have used this approach[26–29]. Fahmy [26] developed regression models to predict theincidence of fuel poverty in England using sample survey data andapplied resultant weights to Census spatial data sets. Similarly,Walker and Day [30] developed a small area fuel poverty risk indexusing environmental and socioeconomic variables via geographi-cal methods finding significant clusters of high and low-risk areas.“The underlying idea is that there are higher probabilities of fuelpoverty in particular areas and/or housing types” [31].

In the U.S., Min et al. [32] applied this approach for spatiallymodeling national residential energy consumption end uses. Com-bining regression models based on national data from the U.S.Energy Information Administration’s (EIA) Residential Energy Con-sumption Survey (RECS) with U.S. Census data, they mapped energyconsumption estimates for space heating, cooling, water heatingand all other electrical uses at the zip code level. Reames [12] usedboth the RECS and Census data to explore racial and socioeconomicdisparities in the spatial distribution of urban heating EUI. Bothstudies found that significant predictors of energy consumptionand EUI included age of housing unit, type of housing unit, numberof rooms, type of heating fuel and household income.

3. Data and methodology

3.1. Description of study area

Detroit (Wayne County) is the largest urban area in the Stateof Michigan and represents nearly 20% of the state population.According to the 2010 decennial census, the county had a total pop-ulation of 1,820,584 residents in 821,693 housing units. Michiganhomes are typically older than homes in other states. Nearly three-quarters of housing stock in Detroit (Wayne County) was builtbefore 1970. Fig. 1 illustrates the distribution of housing stock age,displaying the median year built for block group housing structures.

Socioeconomic characteristics vary in the study area. Detroitexhibits a high and increasing level of residential segregation byincome. The Pew Research on Social and Demographic Trends foundthat the Detroit metropolitan area’s RISI score increased from 43in 1980–54 in 2010 [33].2 Fig. 2 displays the spatial distributionof block group median household incomes, ranging from $6833 to$183,462 per year. Households in the Detroit metropolitan werehit particularly hard during the economic recession and recovery.A survey of Detroit metropolitan area households found that 1 in 2respondents reported experiencing some type of material hardship[34]. While roughly 14% of high-income households fell behind onutility payments, nearly 40% of low-income households reportedbeing behind and were seven times more likely to have a utilityshutoff [34].

culated by summing the share of lower-income households living in a majoritylower-income tract and the share of upper-income households living in a major-ity upper-income tract. The maximum possible RISI score is 200, indicating that100% of lower-income and 100% of upper-income households would be situated ina census tract where most households were in their same income bracket.

D.J. Bednar et al. / Energy and Buildings 143 (2017) 25–34 27

edian

thAcalpu

acgphd

3

hlrceucmiofp

tionships between household energy consumption and variousexogenous variables [39,40,32,12,41]. Statistical models also allowfor capturing consumption variations due to demographic and

Fig. 1. Block group m

he spatial distribution of residents by race/ethnicity. The house-old racial/ethnic composition included 52.3% White, 40.5% Africanmerican and 5.2% Hispanic households. Historically marginalizedommunities of color in Detroit experience higher rates of arrearsnd shutoffs. For instance, African Americans were almost twice asikely as non-African Americans to report being behind on utilitiesayments and more than three times more likely to experience atility service shutoff than non-blacks [34].

Michigan households experience harsher winters increasing theverage household demand for space heating to 55% of total energyonsumption compared to 41% nationally [2]. Consequently, Michi-an households also consume 38% more energy and spend sixercent more than the average U.S. household [2]. Thus, spaceeating is the ideal energy end use for investigating patterns andisparities in consumption and efficiency.

.2. Data

In the absence of detailed individual energy data for everyousehold in the study area, the EIA’s RECS provides household-

evel data for a representative sample of occupied, primaryesidences at the state-level. First conducted in 1978, RECSollected data on energy consumption, annual expenditure,nergy-related behavior, household demographics and housingnit characteristics. Using a multi-stage, area probability design,arefully controlled at specific levels of precision, the 2009 RECSicrodata set (released in 2013) has a sample size of 12,083 hous-

ng units representing the U.S. Census Bureau’s statistical estimatef 113.6 million occupied primary residences [36]. The RECS allowsor state-level analysis with the collection of representative sam-les in 12 states, including Michigan. A sample of 274 Michigan

structure year built.

households were surveyed to represent the state’s 4.5 million occu-pied housing units. Since the scope of this study focuses on annualspace heating, six of the total 274 observations were removed fromthe sample because of missing heating data, resulting in 268 totalobservations for this study.3

Spatial data for modeling and mapping the study area wereobtained from U.S. Census Bureau 2006–2010 American Commu-nity Survey (ACS) [37,38] 5-year estimates. This survey is issuedeach year to provide current information about social and eco-nomic needs of the community. Households are sampled randomlyin each state, including Puerto Rico to provide a representative sam-ple. The census block group was used as the unit of analysis, as themost appropriate spatial resolution for household and housing unitcharacteristics data [12]. A GIS data layer of Wayne County cen-sus block groups was created by clipping the U.S. Census BureauTIGER/Line Shapefile with demographic and economic data fromthe 2006–2010 ACS [37,38] 5-year estimates. Block groups wereonly retained if both population and number of occupied housingunits were greater than zero. Subsequently, 1808 of 1822 blockgroups were included in this analysis.

The RECS microdata set can be used to develop a bottom upstatistical model. These models have been used to explore rela-

3 For a 95 percent confidence interval, a sample size of 246 RECS observations areneeded to prove statistical significance. For geographic domain estimation purposes,base sampling w(YHeat ) or (YEUI).eights were applied to each housing unit. Eachsampling weight value was used as a weighting factor in the weighted regressionmodel.

28 D.J. Bednar et al. / Energy and Buildings 143 (2017) 25–34

edian

sRulast

3h

sRsft

fetc(mhtt2

tt

Fig. 2. Block group m

ocioeconomic characteristics. Similar variables found in both theECS and ACS allow relationships derived from statistical modelssing RECS, known as direct estimates, to be applied to block group

evel ACS spatial data as indirect estimators for constructing small-rea estimates with the assumption that the small area exhibits theame characteristics as the large area [42]. The next section clarifieshis methodological framework.

.3. Methodological framework for estimating block groupeating consumption and efficiency

The goal of this study is to explore residential heating con-umption and efficiency at a geographic domain smaller than theECS microdata, which is collected with adequate precision at atate-level scale. Fig. 4 displays a schematic of the methodologicalramework for estimating heating energy consumption and EUI athe block group level.

The first step uses household and housing unit variables �RECS ,rom the RECS microdata set, specifying two robust regression mod-ls − one to predict residential heating energy consumption andhe other to predict heating EUI (Blue ovals). The second step usesensus data for small area estimations at the block group levelpurple rectangles). Resultant weights, �i, derived from the afore-

entioned robust regression models are multiplied to matchingousehold and housing unit spatial variables (e.g. housing unitype, housing units built in each decade, housing unit heating fuelype, median household income), XCENSUS, from the U.S. Census

006–2010 ACS 5-year estimates.

The objective of the first step is to develop two robust statis-ical regression models that explain the relationship between thewo response variables, heating energy consumption and EUI, with

household income.

the predictor variables, housing unit characteristics (age of home,type of heating fuel, type of home and size of home) and control-ling for household characteristics (household ownership, numberof household members and household income). Dependent vari-ables were natural log values of per-household final consumptionand EUI for heating. The models are formulated as:

ln(YHeat) = ˇ0 +(ˇHousing unit ∗ �RECS

)+

(ˇHousehold ∗ �RECS

), (1)

ln(YEUI) = ˇ0 +(ˇHousing unit ∗ �RECS

)+

(ˇHousehold ∗ �RECS

)(2)

where:YHeat is energy consumption in MJ,YEUI is EUI in MJ/m2,ˇ0 is the regression intercept,ˇHousingUnit is the resultant weight for housing unit characteris-

tics,ˇHousehold is the resultant weight for household characteristics,�RECS is household and housing unit RECS data.The RECS notation is used to differentiate for model creation

in this step, and estimation in the subsequent step using Censusdata. Step one uses resultant weights, ˇi, from the RECS, 2009 datato model energy consumption and EUI. Using the observed datafrom the state of Michigan, a statewide ordinary least squares (OLS)regression model is developed for each response variable, mea-sured in mega joules (MJ) and MJ per square meter per annum. Thegoal of the OLS is to model the relationship between the responseand predictor variables; simply, how housing units and householdcharacteristics influence total heating fuel consumption and EUI.

Total heating consumption is the total annual heating energy con-sumed from all fuel types (i.e. natural gas, electric, fuel oil, liquidpetroleum gas, and kerosene). The EUI is measured as the ratio oftotal heating consumption to total square meters of heated space. A

D.J. Bednar et al. / Energy and Buildings 143 (2017) 25–34 29

Fig. 3. Block group racial/ethnic segregation dot density map.

Fig. 4. Methodological framework for modeling and mapping.

3 nd Bu

lt

�maiunchmpm

l

l

w

c

a

4

casp2rhatFtn

eamuna

e

cm

r

e

et2

0 D.J. Bednar et al. / Energy a

arger EUI value indicates relatively less efficiency when comparedo another housing unit.

Step two applies resultant weights from the regression models,i as weighting factors to corresponding variables in the ACS to esti-ate, then map the median annual heating energy consumption

nd EUI at the block group level in Wayne County. The correspond-ng variables are standardized as the ratio of the number of housingnits in a block group with a certain characteristic to the totalumber of housing units in the block group.4 This is done for eachorresponding variable (age of home, type of heating fuel, type ofome, size of home, household ownership, number of householdembers and household income). These values then become com-

arable with binary variables in the RECS data set. Values are thenapped via GIS to estimate5 heating consumption and EUI:

n(YHeat) = ˆ 0 +(

ˆHousing unit ∗ �Census

)+

(ˆ

Household ∗ �Census

),

(3)

n(YEUI) = ˆ 0 +(

ˆHousing unit ∗ �Census

)+

(ˆ

Household ∗ �Census

)(4)

here:YHeat is estimated energy consumption, in MJYEUI is estimated EUI, in MJ/m2,ˆ 0 is the estimated regression intercept,ˆ

Housing Unit is the estimated sampling weight for housing unitharacteristics,

ˆHousehold is the estimated sampling weight for household char-

cteristics,�Census is household and housing unit Census data.

. Results

The final regression models for estimating annual heatingonsumption and EUI are summarized in Table 1, expresseds natural logs. Model 1, heating consumption, consists of fivetatistically significant variables representing housing unit type,rimary heating fuel and number of household members. Model, heating EUI, consists of six statistically significant variablesepresenting housing unit type, primary heating fuel, number ofousehold members and housing unit size. Both models explained

considerable proportion of the variability in heating consump-ion and EUI (R2 = 0.52, F(18,249) = 15.18, p < 0.001 and R2 = 0.52,(18,249) = 11.09, p < 0.001, respectively). Based on the F-values,he final models’ sample sizes are large enough to make them sig-ificant.

Figs. 5 and 6 display the spatial distribution in quintiles of thestimated mean annual block group heating energy consumptionnd heating EUI, respectively. Red shading represents higher esti-ates, while green shading represents lower estimates. The 14

ninhabited block groups were left uncolored. It is important toote that estimates represent the block group mean rather than

ny specific house [32,43].

Among the 1808 block groups, there was a significant range instimated heating consumption (Fig. 5) values, from a minimum

4 If block group A has 100 homes, and 50 are single family attached, then theorresponding variable for single family attached is 50/100 = 0.5 which would beultiplied by 0.015 (from Table 1).5 From the estimated log values ln(YHeat ) and ln(YEUI) that we obtain from the

egression models, actual estimated energy can be obtain by this equation: (YHeat ) =xp

(RMSE2/2

)· ln(YHeat ); (YEUI(YEUI) = exp

(RMSE2/2

)· ln(YEUI). The scaling value

xp(RMSE2/2) is needed when using a log-linear model because without it we sys-ematically underestimate the expected value of (YHeat ) or (YEUI). (Wooldridge 2006:19). RMSE means root mean square error of each model.

ildings 143 (2017) 25–34

18,658 MJ to a maximum 123,120 MJ. The study area mean heat-ing consumption, 85,107 MJ (SD=16,342 MJ), was lower than thestate mean heating consumption, 131,883 MJ. The 104,451 MJ vari-ation in heating consumption estimates demonstrates that withinthe study area some homes consume a disproportionate amount ofenergy when compared to others. Block groups exhibiting the high-est quintiles of heating consumption primarily surround Detroit onthe east, north and west sides of city.

Estimated heating EUI (Fig. 6) values ranged from a minimum285 MJ/m2 to a maximum 1108 MJ/m2. The study area mean heat-ing EUI, 613 MJ/m2 (SD = 9.8), was lower than the state meanheating EUI, 727 MJ/m2. The 818 MJ/m2 variation in heating EUIestimates demonstrates that within the study area some homesare far less energy efficient than others. Block groups exhibitingthe lowest quintile EUI (shown in green) are located along the west,southwest and east sides of the county, representing homes withhigher levels of energy efficiency. Moderate estimated EUIs, (shownin yellow) are located in the north central portion of the county,while a majority of the higher EUIs, (shown in red) are locatedin the central region of Detroit, indicating lower levels of energyefficiency. This matches areas where houses are older (Fig. 3) andmay suggest that older homes are less energy efficient than newerhomes a few miles outward.

To understand the relationship between heating consumptionand EUI with measures of race/ethnicity and socioeconomic sta-tus, bivariate analysis using pairwise correlation was conducted.Pearson correlations, shown in Table 2, reveal statistically sig-nificant relationships between socioeconomics, education, andhousing tenure with estimated heating consumption (p < 0.001).Heating consumption is positively correlated with block groupswith median household income (0.28) and percent of home-owners (0.56). Furthermore, heating consumption is negativelycorrelated with number of households in poverty (−0.25) and thepercentage of adults without a diploma (−0.07). There are no sig-nificant correlations between heating consumption or EUI withhouseholders above the age of 65. Table 2 also shows statisti-cally significant relationships between socioeconomics, education,race/ethnicity, housing tenure and estimated heating consumptionand EUI (p < 0.001). Contrary to heating consumption, heating EUIis positively correlated with block groups with a higher numberof adults without a high school diploma (0.32), higher number ofhouseholds in poverty (0.32), percentage of African American (0.24)and Hispanic householders (0.16). Heating EUI is negatively corre-lated with median household income (−0.28), percentage of Whitehouseholders (−0.28) and percent of homeowners (−0.38). Thus,census block groups with lower socioeconomics, lower medianhousehold incomes, and higher percentages of African American orHispanic households are more likely to have higher heating EUIs.Simply put, low-income, African American and Hispanic house-holds reside in housing areas where homes consume more and areless energy efficient.

5. Discussion

Results mapped using GIS illustrate inverse spatial disparities inheating consumption and EUI, with higher estimated consumptionin block groups surrounding the central city, while block groupswith higher estimated EUIs are concentrated within the city ofDetroit. The findings also demonstrated that inverse relationshipsexist between the racial and socioeconomic correlations with block

group predicted consumption and EUI. While areas with greaterpercentages of minority households and lower socioeconomic sta-tuses exhibited lower predicted heating consumption, those sameareas exhibited higher EUI, signaling that although low-income,

D.J. Bednar et al. / Energy and Buildings 143 (2017) 25–34 31

Fig. 5. Estimated residential heating consumption in MJ.

Fig. 6. Estimated residential energy use intensity (Efficiency) in MJ/m2.

32 D.J. Bednar et al. / Energy and Buildings 143 (2017) 25–34

Table 1OLS regression models for small-scale heating consumption and EUI estimation.

Categories Model 1:Heating Consumption(MJ)

Model 2:Energy Use Intensity(MJ/m2)

Type of Housing Robust S.E. Robust S.E.Apt 2–4 0.1431 0.1317 0.6728*** 0.1864Apt 5> −0.2989* 0.1401 0.2987* 0.1511Mobile Home 0.5173* 0.2361 0.1090 0.2271Single Family Detached Reference ReferenceSingle Family Attached 0.01531 0.1402 0.0948 0.1809

Decade ConstructedBefore 1950s 0.3317 0.1739 0.328 0.171950s 0.3223 0.1802 0.3521 0.17861960s 0.0681 0.1769 0.1126 0.18481970s −0.0026 0.1832 −0.0159 0.1881980s 0.0383 0.1693 0.0843 0.18081990s −0.0124 0.216 −0.1232 0.21252000s Reference Reference

Primary Heat (MJ)Natural Gas Reference ReferencePropane 0.0138 0.0855 −0.098 0.1055Electricity −1.627*** 0.1404 −1.381*** 0.1677Wood −1.170 0.6978 −1.198 0.6732Fuel Oil Heat −0.6926* 0.270 −0.6823** 0.2061

Control VariablesHousehold Income ($) 0.0228 0.026 −0.0012 0.0259No. Household Members −0.0506* 0.0256 −0.0619* 0.0266Home Ownership (own = 1) 0.0806 0.0853 −0.01029 0.0874

Total No. of rooms 0.0203 0.0279 −0.1048*** 0.0254Model StatisticsIntercept, ˇ0 10.87375 0.2568 4.269 0.248N 268 – 268 –F (18,249) – 15.18 – 11.09Adjusted R2 0.5242 – 0.5183 –RMSE – 0.514 – 0.574

* Significance p < 0.05.** Significance p < 0.01.

*** Significance p < 0.001.

Table 2Pairwise Correlation of Estimated Heating Energy Consumption and Energy Use Intensity.

Category Description Pearson’s Correlation

Heating Consumption Heating Intensity

Socioeconomic Status Median Household income 0.28*** −0.48***

Percent households below poverty level −0.25*** 0.32***

Education Percent Population with Less Than High School Diploma −0.07** 0.31***

Age Percent Households with Householder aged 65+ 0.01 0.02Race/Ethnicity Percent White Householders 0.23 −0.28***

Percent African American Householders −0.01 0.24***

Percent Hispanic Householders 0.02 0.16***

Housing Tenure Percent Owners 0.56*** −0.38***

mm

odmasUascfpc

** Significance p < 0.01.*** Significance p < 0.001.

inority households on average consume less energy, they areore likely to live in less efficient housing.

Studying cities like Detroit is important because they often havelder housing stock central to the city with much newer, suburbanevelopments outside the city. As shown, householders occupyinguch older housing stock are at a greater risk for increased demand

nd a greater need for energy assistance programs. Although thistudy is focused in the south-east region of Michigan within thenited States, this study could be replicated in other urban areas,s well as other countries using a similar household energy con-umption survey (i.e. Zheng [44]; ODYSSEE MURE Project) and thatountry’s census data. The significance of the results presented callor an integrated approach that tackles fuel poverty from both a

hysical and policy standpoint − evaluating building energy effi-iency and energy assistance programs.

5.1. Policy implications

Energy assistance programs provide eligible householders withmonetary or housing unit efficiency upgrade support. The federallyfunded Low Income Heating Energy Affordability Program (LIHEAP)provides energy assistance to residents whom are unable to affordtheir high utility bills. Identifying concentrated areas of high EUIand energy burden is still a concern given the aforementionedsupport from government. LIHEAP eligibility primarily dependson income; however, many qualified householders do not receiveenergy assistance. While attenuating exorbitant utility bills pro-vides temporary relief for some householders, it perpetuates fuelpoverty by not combatting a root cause, energy inefficiency.

The U.S. Department of Energy Weatherization Assistance Pro-gram’s (WAP) purpose, as established by law, “provides low-tono-cost energy efficiency improvements of dwellings owned or

nd Bu

oeipeiLgEtounoeeyaaiOfiapirhia

prtrdct

otaitst[

5

uoiscrgemFs

6

I

D.J. Bednar et al. / Energy a

ccupied by low-income persons, reduces their total residentialxpenditures, and improves their health and safety, especially low-ncome persons who are particularly vulnerable such as the elderly,erson with disabilities, families with children, high residentialnergy users, and households with high energy burden” [45]. WAPs monitored by the Department of Energy’s Oak Ridge Nationalaboratory (ORNL). ORNL provides technical support to the pro-ram and conducts the evaluations. Led by ORNL, the Department ofnergy sponsored two major national evaluations: The Retrospec-ive Evaluation (covering Program Year 2008, which is reflectivef a typical year in WAP operations) and the Recovery Act Eval-ation (covering Program Year 2010, providing insight to theational effort of job creation and economic recovery as a partf the American Recovery and Reinvestment Act of 2009 [Recov-ry Act]) were multiyear, peer-reviewed and statistically robustfforts. The former was performed to provide a cost-benefit anal-sis of WAP services for varying housing unit types and locationscross the country. Additionally, to assess program administrationnd to provide a comprehensive overview of the program, includ-ng information on its clients, housing stock and service providersRNL, 2014. Effective and optimal funding of the system is veri-ed through “whole-house” weatherization approaches via energyudits and the three-pronged WAP funding allocation formula:ercent of low-income population, climatic conditions and approx-

mate residential energy burden. Challenges of WAP presentedevolve around maintaining and improving work quality, handlingealth and safety issues discovered in homes and meeting a grow-

ng demand for program services. Further, the Recovery Act did notddress renewable energy measures average costs per home.

Though LIHEAP and WAP help mitigate energy burdens, theserograms do not permit the use of sustainable energy, such asenewable energy for heating and cooling. Renewable energy sys-ems have proven beneficial for energy generation with respect toetrofits [46–48]. There is an opportunity for growth that intro-uces renewables as a conduit for greater efficiency; however, aommunity based approach would be more fruitful for effectiveargeting.

Community-based energy programs have shown success forvercoming various barriers and increasing participation inhe adoption of energy technologies [43]. A community-basedpproach to energy efficiency that targets low-income and minor-ty communities recognizes the unique characteristics and needs ofhe community and can better foster equity and justice over typicalelf-referral, broad- based program development and implemen-ation which relies on a homogeneous view of energy users49,43,30].

.2. Limitations

As with all research, this study is limited in its scope to fullynderstand individual households in fuel poverty. Informationbtained from this data is often not precise enough to identify

ndividual households; rather, only census block groups at risk ofuffering from fuel poverty. Although, some homes that are notonsidered fuel poor may become integrated spatially with sur-ounding ones that are, this study provides a model of mean blockroup estimates to inform policy and program targeting whilexploring relationships with race/ethnicity and class. Specific infor-ation about individual household utility bills is not accessible.

urther, the influence of behavior on disparities in energy con-umption or efficiency are not observed in these models.

. Conclusion

This study used publically available data from the U.S. Energynformation Administration’s Residential Energy Consumption Sur-

ildings 143 (2017) 25–34 33

vey (RECS) and the U.S. Census Bureau’s American CommunitySurvey (ACS), bottom-up modeling, and small-area estimationtechniques to predict mean annual heating consumption andenergy use intensity (EUI), an energy efficiency proxy, for censusblock groups in Detroit (Wayne County), Michigan. This study’s rel-evance provides a best estimate of areas where householders mayexperience the greatest threat of fuel poverty. The key findings ofthe study illustrate inverse spatial disparities in heating consump-tion and EUI throughout Detroit (Wayne County), Michigan. Inverserelationships were also found between the racial and socioeco-nomic correlations with block group predicted consumption andEUI.

Modeling both heating consumption and efficiency provides auseful tool that may assist policymakers, energy conservation andefficiency program administrators and retrofit installers developmore effective targeting strategies. Combining consumption andefficiency information with an understanding of the racial andsocioeconomic context of neighborhoods may also improve pro-gram implementation effectiveness.

Using spatial proximity as a guide to identifying fuel poor house-holders eliminates onerous applications to determine eligibilityand provides a quicker and more robust response to householdersin need. Furthermore, there is a need to understand the cul-tural/racial differences within identified neighborhoods. Simplycreating energy assistance programs without effective marketing,maintains the energy divide, leaving many in fuel poverty. To over-come cultural and social barriers, community-based approacheswould enable more access to help that is readily available. Futureresearch should pursue a more granular level of understanding,such as, incorporating individual parcel data. Additionally, spatiallymodelling of energy burdens would provide a more holistic viewof residential energy assistance demands. With this informationin hand, program administrators could target local churches, com-munity centers and neighborhood groups to more effectively andefficiently assist those with the greatest need.

Addressing fuel poverty and energy consumption more broadly,requires an integrated approach to identify the specific energyneeds of communities. The modeling framework presented inthis study is one approach to understand those needs both visu-ally and statistically. Moreover, this research unpacks disparitiesin consumption and efficiency concluding that one-size-fits-allapproaches to conservation and efficiency are not appropriate forall energy users in an urban area.

References

[1] J. Siebens, Extended Measures of Well-Being: Living Conditions in the UnitedStates: 2011, United States Census Bureau, 2011.

[2] U.S. Energy Information Administration (EIA), (2013a). Residential EnergyConsumption Survey, 2009.

[3] B. Boardman, Fuel Poverty: From Cold Homes to Affordable Warmth,Belhaven Press, London, 1991.

[4] D. Hernandez, Energy insecurity: a framework for understanding energy, thebuilt environment and health among vulnerable populations in the context ofclimate change, Am. J. Public Health 104 (2013) 3.

[5] C. Goodacre, S. Sharples, P. Smith, Integrating energy efficiency with the socialagenda in sustainability, Energy Build. 34 (1) (2002) 53–61, http://dx.doi.org/10.1016/S0378-7788(01)00077-9.

[6] S.H. Hong, T. Oreszczyn, I. Ridley, The impact of energy efficient refurbishmenton the space heating fuel consumption in English dwellings, Energy Build. 38(10) (2006) 1171–1181, http://dx.doi.org/10.1016/j.enbuild.2006.01.007.

[7] H. Tonn, J. Rose, S. Svendsen, Energy-efficient houses built according to theenergy performance requirements introduced in Denmark in 2006? EnergyBuild. 39 (10) (2014) 1123–1130.

[8] H. Tonn, J. Rose, S. Svendsen, Energy-efficient houses built according to theenergy performance requirements introduced in Denmark in 2006? Energy

Build. 39 (10) (2015) 1123–1130.

[9] S. Heiple, D.J. Sailor, Using building energy simulation and geospatialmodeling techniques to determine high resolution building sector energyconsumption profiles, Energy Build. 40 (8) (2008) 1426–1436, http://dx.doi.org/10.1016/j.enbuild.2008.01.005.

3 nd Bu

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

468–492, http://dx.doi.org/10.1525/sp.1995.42.4.03x0128x.

4 D.J. Bednar et al. / Energy a

10] B. Howard, L. Parshall, J. Thompson, S. Hammer, J. Dickinson, V. Modi, Spatialdistribution of urban building energy consumption by end use, Energy Build.45 (2012) 141–151, http://dx.doi.org/10.1016/j.enbuild.2011.10.061.

11] M. Santamouris, K. Kapsis, D. Korres, I. Livada, C. Pavlou, M.N.Assimakopoulos, On the relation between the energy and socialcharacteristics of the residential sector, Energy Build. 39 (8) (2007) 893–905,http://dx.doi.org/10.1016/j.enbuild.2006.11.001.

12] T.G. Reames, Targeting energy justice: exploring spatial, racial/ethnic andsocioeconomic disparities in urban residential heating energy efficiency,Energy Policy 97 (2016) 549–558, http://dx.doi.org/10.1016/j.enpol.2016.07.048.

13] L. Adua, To cool a sweltering earth: does energy efficiency improvement offsetthe climate impacts of lifestyle? Energy Policy 38 (10) (2010) 5719–5732.

14] Horrace Herring, Is energy efficiency environmentally friendly? EnergyEnviron. 11 (3) (2000) 313–325.

15] Willett Kempton, Laura Montgomery, Folk quantification of energy, Energy 7(10) (1982) 817–827.

16] M.D. Levine, J.G. Koomey, J.E. McMahon, A.H. Sanstad, E. Hirst, Energyefficiency policy and market failure, Annu. Rev. Energy Environ. 20 (1995)535–555.

17] L. Lutzenhiser, Social and behavioural aspects of energy use, Annu. Rev.Energy Environ. 18 (1993) 247–289, http://dx.doi.org/10.1146/annurev.energy.18.1.247.

18] M. Moezzi, Decoupling energy efficiency from energy consumption, EnergyEnviron. 11 (2000) 521–537.

19] Andrew Rudin, Efficiency and conservation: an interview with andy rudin,Energy Environ. 15 (6) (2004) 1085–1092.

20] Lee Schipper, Life-styles and Energy: A New Perspective, Lawrence BerkeleyLab., Berkeley, CA, 1991.

21] L. Schipper, M. Grubb, On the rebound? Feedback between energy intensitiesand energy uses in IEA countries, Energy Policy 28 (6–7) (2000) 367–388,http://dx.doi.org/10.1016/S0301-4215(00)00018-5.

22] C. Starr, M.F. Searl, S. Alpert, Energy sources: a realistic outlook, Science (NewYork, N.Y.) 256 (5059) (1992) 981–987, http://dx.doi.org/10.1126/science.256.5059.981.

23] H. Tommerup, J. Rose, S. Svendsen, Energy-efficient houses built according tothe energy performance requirements introduced in Denmark in 2006?Energy Build. 39 (10) (2007) 1123–1130.

24] Lee Schipper, Bartlett Sarita, Hawk Dianne, Vine Edward, Linking lifestyle 111and energy use: a matter of time? Ann. Rev. Energy 14 (1989) 273–320.

25] Paul C. Stern, Blind spots in policy analysis: what economics doesn’t say aboutenergy use, J. Policy Anal. Manage. 5 (2) (1986) 200–227.

26] E. Fahmy, D. Gordon, D. Patsios, Predicting fuel poverty at a small-area level inEngland, Energy Policy 39 (7) (2011) 4370–4377, http://dx.doi.org/10.1016/j.enpol.2011.04.057.

27] M. Santamouris, J.A. Paravantis, D. Founda, D. Kolokotsa, P. Michalakakou,A.M. Papadopoulos, E. Servou, Financial crisis and energy consumption: ahousehold survey in Greece, Energy Build. 65 (2013) 477–487, http://dx.doi.org/10.1016/j.enbuild.2013.06.024.

28] R. Walker, P. McKenzie, C. Liddell, C. Morris, Estimating fuel poverty athousehold level: an integrated approach, Energy Build. 80 (2014) 469–479,http://dx.doi.org/10.1016/j.enbuild.2014.06.004.

29] R. Walker, P. McKenzie, C. Liddell, C. Morris, Area-based targeting of fuel

poverty in Northern Ireland: an evidenced-based approach, Appl. Geogr.(2012) 639–649, http://dx.doi.org/10.1016/j.apgeog.2012.04.002.

30] G. Walker, R. Day, Fuel poverty as injustice: integrating distribution,recognition and procedure in the struggle for affordable warmth, EnergyPolicy 49 (2012) 69–75, http://dx.doi.org/10.1016/j.enpol.2012.01.044.

[

ildings 143 (2017) 25–34

31] U. Dubois, From targeting to implementation: the role of identification of fuelpoor households, Energy Policy 49 (2012) 107–115, http://dx.doi.org/10.1016/j.enpol.2011.11.087.

32] J. Min, Z. Hausfather, Q.F. Lin, A high-resolution statistical model of residentialenergy end use characteristics for the United States, J. Ind. Ecol. 14 (5) (2010)791–807, http://dx.doi.org/10.1111/j.1530-9290.2010.00279.x.

33] R. Fry, P. Taylor, The rise of residentail segregation by income, in: Social &Demographic Trends, Pew Research Center, 2012Http://www.pewsocialtrends.org/2012/08/01/the-rise-of-residential-segregation-by-income/.

34] A. Gould-Werth, K. Seefeldt, Material Hardships During the Great Recession:Findings from the Michigan Recession and Recovery Study, National PovertyCenter Policy Brief #35, 2012.

35] J. Logan, B. Stults. The Persistence of Segregation in the Metropolis: NewFindings from the 2010 Census. US2010 Project, 2011.

36] U.S. Energy Information Administration (EIA), Residential EnergyConsumption Survey, 2013 https://www.eia.gov/consumption/residential/methodology/2009/pdf/techdoc-summary010413.pdf.

37] U.S. Census Bureau. American Community Survey, American CommunitySurvey 5-Year Estimates, 2010.

38] U.S. Census Bureau. State & county Quickfacts: Wayne County, MI, 2010.Retrieved January 21, 2016, from http://quickfacts.census.gov.

39] A.T. Booth, R. Choudhary, Decision making under uncertainty in the retrofitanalysis of the UK housing stock: implications for the Green Deal, EnergyBuild. 64 (2013) 292–308, http://dx.doi.org/10.1016/j.enbuild.2013.05.014.

40] R. Ewing, F. Rong, The impact of urban form on U.S. residential energy use,Hous. Policy Debate 19 (1) (2008) 1–30, http://dx.doi.org/10.1080/10511482.2008.9521624.

41] G.K.F. Tso, K.K.W. Yau, Predicting electricity energy consumption: acomparison of regression analysis, decision tree and neural networks, Energy32 (9) (2007) 1761–1768, http://dx.doi.org/10.1016/j.energy.2006.11.010.

42] J.N.K. Rao, I. Molina, Small area estimation, in: Wiley Series in SurveyMethodology, 2nd. ed., Wiley, 2015, http://dx.doi.org/10.1002/9781118735855.

43] T.G. Reames, A community-based approach to low- income residential energyefficiency participation barriers, Local Environ. 21 (12) (2016) 1449–1466,http://dx.doi.org/10.1080/13549839.2015.1136995.

44] X. Zheng, C. Wei, P. Qin, J. Guo, Y. Yu, F. Song, Z.1 Chen, Characteristics ofresidential energy consumption in China: findings from a household survey,Energy Policy 75 (2014) 126–135, http://dx.doi.org/10.1016/j.enpol.2014.07.016.

45] U.S. Department of Energy (DOE). Federal Register. Weatherization AssistancePrograms for Low Income Persons, Vol. 65, No. 237, 2006, pp. 77210–77219.

46] J.A. Lepore, S. Shore, N. Lior, Retrofit of urban housing for solar energyconversion, Hous. Sci. 2 (6) (1978) 483–498.

47] N. Lior, First year performance monitoring of an urban row house retrofittedto solar heating, in: Proc. Annu. Meet.-Am. Sect. Int. Sol. Energy Soc., (UnitedStates), 1980, 3(CONF-800604-P2).

48] N. Lior, Retrofit for solar heating and cooling, in: Advances in Solar Energy,Springer, US, 1989, pp. 360–401.

49] L. Higgins, L. Lutzenhiser, Ceremonial equity: low-Income energy assistanceand the failure of socio-environmental policy, Soc. Probl. 42 (4) (1995)

52] U.S. Department of Energy (DOEa). Office of Energy Efficiency & RenewableEnergy. Energy Intensity Indicators: Efficiency vs. Intensity. https://energy.gov/eere/analysis/energy-intensity-indicators-efficiency-vs-intensity.Accessed November 2015.